Elsevier

Energy Economics

Volume 30, Issue 5, September 2008, Pages 2697-2704
Energy Economics

The influence of temperature on spike probability in day-ahead power prices

https://doi.org/10.1016/j.eneco.2008.05.007Get rights and content

Abstract

It is well known that day-ahead prices in power markets exhibit spikes and time-varying volatility. Spikes and extremely high volatility are the results of (short-term) frictions in demand and/or supply conditions. It is known that information on load or the reserve margin help to forecast spikes. However, these variables are not (timely) available for every market participant and this paper suggests to use temperature as a proxy. Interpreting the results from several switching-regimes models, the paper shows that the probability of spike occurrence increases when temperature deviates substantially from mean temperature levels.

Introduction

Prices in day-ahead electricity markets exhibit frequently periods with higher volatility than normal and spikes. These occur due to frictions in demand and/or supply to which electricity producers cannot respond flexible enough. Sometimes these frictions results in huge price spikes as, for example, in the Dutch APX market in August 2003. A heat wave was causing the temperature of water in rivers to reach such levels that the Dutch government decided to restrict the outflow of cooling water from, among others, energy producers into the rivers, thereby effectively limiting the installed production capacity. Prices in the day-ahead market spiked, reaching levels over 1000 euro for 1 MW of electricity delivered in the peak hours on August 11 and 13, whereas the average price in the peak hours was approximately 49 euro over the 20 days before August 11. This example of a supply driven cause of a multi-day price spike shows that the existence of these spikes cannot be ignored as these spikes have an impact on the amount of market risk and credit risk that companies face. For instance, a power distribution company that purchases a part of their client volume on the day-ahead market will be confronted with a huge increase in their purchasing price of power as a result of a spike. If the company is committed to sell against fixed prices, these spikes will dramatically impact the profits and losses of the company. Not only the profitability itself is at stake (market risk), but also counter-parties and other stakeholders will observe the reduction in profitability and perceive the company as more risky (credit risk). In order to protect their positions, the counter-parties might ask for more collateral, leading to an additional cash-outflow or – at least – a reduced amount of liquid assets. Therefore, if companies purchase a part of their electricity needs on the day-ahead markets, they need to manage the market and credit risk that they face from spikes and high volatility periods. Models, that describe the behavior of day-ahead prices in power markets, help to measure risk, to forecasts cash-flow sensitivity and to valuate derivative contracts on day-ahead delivery such as options, swaps and forwards. It is of crucial importance that these models capture the spike and volatility dynamics of day-ahead electricity prices.

Over the past years, models have been introduced that describe the behavior of day-ahead prices in power markets. Bunn and Karakatsani (2003) provide an excellent review of this literature. It is commonly known that day-ahead prices exhibit mean-reversion, seasonality, spikes and time-varying volatility. Focusing on spikes, these were initially modeled as a jump diffusion process. In this framework, spikes can occur at all times and are an integral part of the model itself. As a result, the mean-reversion parameter in the model reflects both the amount of mean-reversion in normal markets and the mean-reversion after a spike has occurred. Deng (1998), Ethier and Mount (1999), and Huisman and Mahieu (2003) formulate regime-switching models to capture spike behavior. Basically, they observe that spikes occur in situations where the market is not in a normal state due to a – short term – frictions in supply and demand conditions. Therefore, they model day-ahead prices assuming that the market can be in one of two regimes: a normal and a non-normal regime. In each regime, day-ahead prices are modeled according to the assumption of market conditions under which the regime applies. In the normal regime, day-ahead prices are typically modeled as a mean-reversion process and in the non-normal regime, day-ahead prices are typically modeled as a jump. In addition, a Markov process governs the daily transition from one regime to another.

A drawback of these models is that the probability with which a spike occurs is constant over time. That is, the probability of a spike is the same in summer and winter months, in weekdays and weekend, for all weather conditions, and for all levels of reserve capacity. This is not realistic as spikes occur as a result of shocks in demand and supply and these shocks may be caused by some event. Mount et al. (2006) observe this and propose a regime switching model in which the probability of spike occurrence is time-varying. More specifically, they argue and show that the probability of a spike depends on the reserve margin. The lower the reserve margin (the difference between available capacity and capacity in use), the higher the probability on a spike is, as there is probably less capacity available to compensate for shocks in supply and demand in periods with low reserve margins. The authors show that this specification better predicts day-ahead prices as the probability of a spike now depends on actual market conditions.

The motivation for this paper comes from the observation that Mount et al. (2006) make in their conclusions. They state, correctly, that in order to predict day-ahead prices effectively, one needs to have access to accurate information about reserve margins. They fit their model to day-ahead PJM prices and for that market historical information on load and reserve margin was available. However, this is not the case in all markets. Furthermore, if information on reserve margins is available, it might not be easily accessible to every market participant and it might not be available on time. This makes their approach relatively difficult to implement. The purpose of this paper is to partially deal with this problem by introducing temperature as a variable that might influence the probability on a spike as a proxy for reserve margin. As described in the example from the APX market, power consumption depends on temperature (heating in winters, air-conditioning in summers) and shocks in supply or demand are related to shocks in temperature (if today is warmer than expected, consumers use more power than expected for air-conditioning and a short-term shortage, due to planning problems, might occur). The advantage of using temperature information is that data is transparent and more widely available. It may therefore replace reserve margin as a forecasting variable for those markets where information on reserve margin is not or not accurately available. Another advantage of using temperature instead of reserve margins is that risk from changes in day-ahead prices can be hedged more effectively with weather derivatives as these are being traded worldwide and reserve margin derivatives do not exist directly. The impact of temperature and more general weather variables has been studied by Knittel and Roberts (2005), Kosater (2006), and Huurman et al. (2008). All find that weather variables significantly influence power prices and the variance of power prices. This paper extends these studies by examining the impact of temperature on regime transition probabilities in addition to the general price level as a potential variable to forecast spikes or high volatility periods. Therefore, different regime-switching models are discussed: a general one without temperature dependency, one in which temperature affects the general price level and one in which temperature affects also transition probabilities. The parameters of these models are then estimated using data from the Dutch APX market and interpreted in order to examine the actual impact of temperature on day-ahead prices.

Section snippets

A temperature dependent regime-switching model

This section develops three regime switching models for day-ahead electricity prices. The first regime switching model is a general one and serves as a benchmark for the other model. It does not depend on temperature variables. In the second model, temperature is assumed to influence the price level of electricity. The third model extends the second model with having temperature influencing the transition probabilities between the regimes. The construction of the models is based on Huisman and

Data and estimation issues

The data consists of average prices in peak hours on the Dutch APX market between January 1st, 2003 and February 4th, 2008 (1861 observations). Peak hours are assumed to be from 08:00 h through 20:00 h. The temperature data is obtained from the Royal Netherlands Meteorological Institute (KNMI) and can be obtained from the www.knmi.nl website. The temperature reflects the average daily temperature observed in the center of the Netherlands. The reader is referred to Mount et al. (2006) and

Results

The estimates for the parameters in the above models are listed in Table 1.

Firstly, focus on the results for Model(1). All parameter estimates are significantly different from zero. The mean log price µ1 equals 3.993. In weekends, the mean log price is lower by 0.480 (β1). In the normal regime, the speed of mean-reversion is 0.157 and the volatility is 0.227. In regime 2, the regime reflecting frictions in demand and supply, the volatility (0.635) is substantially higher. In addition, the mean

Concluding remarks

The goal of this paper is to extends the findings of Mount et al. (2006) who show that the probability of a spike increases in periods with low reserve margins. This intuitive idea may not be applicable to all markets as information on demand, load and capacity is not transparent and available to all agents in the market in any country. As this paper shows, temperature can be used as a proxy variable that replaces reserve margin under the assumption that temperature directly influences demand

References (12)

There are more references available in the full text version of this article.

Cited by (42)

  • An analysis of price spikes and deviations in the deregulated Turkish power market

    2019, Energy Strategy Reviews
    Citation Excerpt :

    On the other hand, studies that focus on price spikes are limited. Lu et al. (2005) and Zhao et al. (2007) use a data mining based approach [11,12]; Amjady and Keynia (2010) propose a hybrid data model that includes wavelet and time domain variables [13]; Amjady and Keynia (2011) also use probabilistic neural network and hybrid neuro-evolutionary systems [14]; Huisman (2008) analyzes the effect of temperature on price spikes [15]; Mount et al. (2006) propose a stochastic regime-switching model to represent price volatilities and spikes [16]; Guan et al. (2001) analyze the historical bidding behavior in the California day-ahead market and analyze how bidding decisions increase prices and cause spikes [17]. In this paper, we use price and load data from the Turkish power market for the years 2012–2015.

  • An options pricing approach to ramping rate restrictions at hydro power plants

    2016, Journal of Economic Dynamics and Control
    Citation Excerpt :

    De Jong (2006) modifies the spike regime by using an autoregressive process with Poisson jumps. Further extensions are given by Mount et al. (2006) who allow transition probabilities to depend on the reserve margin and Huisman (2008) who suggests temperature dependent transition probabilities. In two recent studies, Janczura and Weron (2009) and Janczura and Weron (2010) introduce median-shifted spike regime distributions with heteroskedastic dynamics for the base regime.

View all citing articles on Scopus

The author thanks two anonymous referees and Alexander Boogert for their helpful comments.

View full text